Cybersicherheit: Hacker dringen offenbar mit Anthropic-KI in OpenAI ein
Sicherheitsforscher haben einem Medienbericht zufolge das KI-Modell Claude genutzt, um sich in OpenAI zu hacken. Der Vorfall zeigt einmal mehr die Risiken von KI in der Cybersicherheit.
Independent security researchers allegedly gained access to OpenAI's systems using Anthropic's AI model Claude, according to the "Wall Street Journal". The hackers participated in OpenAI's bug-hunting program, which invites researchers to find security flaws in the system. Hacktron AI utilized a security vulnerability in the Discourse discussion forum service that OpenAI uses for its community forum.
By leveraging Claude Opus 5, they wrote cyber attack code to exploit this vulnerability, ultimately gaining access to a Discourse server and accessible access tokens. These stolen access tokens allowed the researchers to unexpectedly infiltrate several OpenAI employees' ChatGPT accounts and the company's internal GitHub system. Consequently, they were able to view the company's software repository, the source code system "Monorepo", which reportedly contains OpenAI's algorithmic secrets.
The researchers reported their findings to the company and received $6,500 in compensation from OpenAI.
Written by urgent.news from Handelsblatt's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
